Mobile Analytics When One Person Spans App and Web

Mobile is charged for the research and desktop is credited with the sale. How to resolve identity across devices, the three honest methods, and what to measure once the person is whole.

KISSmetrics Editorial

|11 min read

Mobile converts worse than desktop in almost every dataset, and a large part of that gap is an artefact. Research happens on a phone, purchase happens on a laptop, and two separate anonymous visitors are recorded where there was one person. Mobile is charged for the research and desktop is credited with the sale.

The gap is not entirely artefact. Mobile checkout genuinely is harder, forms genuinely are worse on a small screen, and some of the difference is real. But you cannot size the real part until you have removed the counting part, and most teams have been optimising against the combined number for years.

The newsletter

Join our KISS newsletter

One short read a week on what actually moves revenue, in a free email. Read by 10,000+ operators and founders.

No spam. Unsubscribe in one click.

I.The mobile conversion gap is mostly a counting error

A journey split across devices is recorded as two failures instead of one success.

A.What the default reporting does

Analytics identity is per-browser. A cookie on a phone and a cookie on a laptop are two different people, and nothing about the underlying human is available to connect them unless you provide it.

So the standard high-consideration journey, which is: discover on mobile from social, research on mobile, return on desktop, buy, produces one mobile visitor who browsed and abandoned, and one desktop visitor who arrived direct and purchased. Mobile's conversion rate goes down. Desktop's goes up. Direct traffic gets credited for a sale that social started.

One purchase, recorded as two journeys

Funnels view
Mobile: arrived from social
100100%
Mobile: viewed product
100100%
Mobile: abandoned
100100%
Desktop: arrived direct
100100%
Desktop: purchased
100100%
Illustrative. Every row is the same person. Two of the three channel and device conclusions a reader would draw from this are wrong.

B.The part of the gap that is real

Having said that, do not swing to the opposite error. Mobile checkout is genuinely harder, and the difficulties are specific rather than general.

Forms are the main one. Every additional field costs disproportionately more on a phone, address entry is slow, and card entry without a stored wallet is genuinely unpleasant. Error recovery is worse, because a validation message can push the field off-screen. And interruption rates are higher: a phone is used in more contexts where the session simply ends. Our guide to form optimisation covers most of the fixable half.

The point of resolving identity is to find out how much of your gap is which. A team that discovers 70% of its mobile abandonment finishes on desktop has a measurement problem. A team that discovers 15% does has a checkout problem, and should go and fix the form.

II.Identity is the whole problem, and it has three honest answers

Every cross-device claim rests on how the identities were joined. Ask that first, always.

A.Deterministic on login

The strongest method and the most limited. When someone authenticates, you know exactly who they are, and every event before and after can be attached to that identity.

The limitation is obvious: it only covers people who log in, which excludes the entire pre-purchase journey for most e-commerce and a large share of consideration for most SaaS. It is the ground truth you validate other methods against rather than the method itself.

B.Deterministic on any shared identifier

Broader, and the one that does most of the real work. Any moment a person supplies something stable, you can join. An email address in a newsletter form, a click from an email you sent that carries an identifier, an order confirmation, a support conversation.

The technique is to identify at every one of these moments rather than only at login. Most implementations identify once, at signup, and miss the four earlier opportunities that would have connected the anonymous research to the eventual account.

The three methods, honestly rated

Identity view
MethodAccuracyCoverageUse it for
LoginExactLow, post-purchase mostlyValidating the others
Shared identifierExact when presentModerate, and improvableThe actual join
ProbabilisticUnverifiableHighDirectional sizing at most
Coverage is the column that decides whether an analysis is possible. Accuracy is the column that decides whether it is trustworthy. You need both, and probabilistic methods buy the first with the second.

C.Probabilistic, and why to be careful

Probabilistic matching infers that two devices belong to one person from IP, timing, behaviour and fingerprint signals. It has high coverage and it cannot be verified from inside itself, which is the problem.

It also produces confident errors in exactly the situations that matter: shared home networks, office IPs, and families. Two colleagues researching the same product from one office become one person with an implausible journey. If you use it, use it for sizing an effect and never for a customer-level decision, and be aware that the privacy posture around fingerprinting continues to tighten.

III.What to measure once the person is whole

The mobile question changes shape entirely once the journey is one record.

A.Device belongs on the event, not on the person

The most common modelling error after identity itself: storing device as an attribute of the person. It is not one. It is an attribute of a moment.

A person is not a mobile user. They were on mobile at 8pm and on a laptop at 10am, and treating device as a permanent label means every cross-device analysis you attempt will be forced into a category that does not fit. Store it per event and let the journey show its own shape.

B.The three metrics worth having

%
of purchases that were cross-device
the size of the artefact
the dominant hand-off
which device starts, which finishes
time between devices
how long consideration really takes
What to report once identity is resolved, instead of a mobile conversion rate.

The cross-device share tells you how much of your reporting was wrong. The hand-off direction tells you what each device is actually for in your business, which is usually more specific than the assumption. And the elapsed time tells you how long your attribution window needs to be before it stops truncating real journeys.

That last one has immediate consequences. If half your purchases involve a gap of more than seven days, then any attribution model with a seven-day window is discarding half your first touches. See first-touch versus last-touch attribution.

C.Getting the join without building it

Everything here rests on a person-level identity model, and retrofitting one is expensive because it touches ingestion rather than reporting.

Kissmetrics resolves every event to a person by design and stitches devices whenever an identify call supplies a shared identifier, which means the cross-device journey is the default view rather than a project. Device sits on the event where it belongs, so a funnel can span a phone and a laptop without any of the steps being lost to a second identity.

The honest limit is the same one everybody has: it can only join what a person gives you. Nothing resolves a journey where the visitor never supplied a single shared identifier anywhere, and no vendor claiming otherwise is doing it deterministically.

Verdict

Mobile does convert worse. Some of that is real and fixable in the checkout, and some of it is a person being counted twice because the analytics has no way to know they are the same human.

Resolve identity first, then measure the split. Until you know what share of mobile abandonment finishes somewhere else, every mobile optimisation decision is being made against a number that mixes a design problem with a counting problem, and the two need completely different work.

Continue Reading

mobile analyticscross-device trackingidentity resolutionmobile conversion ratedevice stitchingapp and web analytics
KISSmetrics

Build your business intelligence layer for free.

KISSmetrics resolves events to a person and stitches devices whenever an identifier appears, so a journey that starts on a phone and finishes on a laptop is one record instead of two failures.